Abstract Soil moisture across South America exhibits coherent spatial patterns shaped by upwind terrestrial conditions, yet their downwind propagation remains poorly quantified. We develop an observation-driven framework integrating atmospheric transport trajectories with deep le…
Behavior prior reinforcement learning (BPRL) has emerged as a promising paradigm to improve sample efficiency in online reinforcement learning (RL) by leveraging policy priors derived from offline demonstrations. However, most existing BPRL methods rely on static offline datasets…